Skip to content
PyTorch implementation of the paper
Jupyter Notebook Python
Branch: master
Clone or download
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
__pycache__ updated method to plot predicted depthmap Jan 14, 2019
data added other files Dec 30, 2018
extras added other files Dec 30, 2018
results added code to test on sample image, inference on cpu, tested on pytor… Dec 30, 2018
.gitignore updated method to plot predicted depthmap Jan 14, 2019
IMG_4120.jpg Add files via upload Dec 30, 2018
IMG_4121.jpg Add files via upload Dec 30, 2018
IMG_4122.jpg Add files via upload Dec 30, 2018
LICENSE Initial commit Dec 30, 2018
README.md Update README.md Dec 30, 2018
custom_transforms.py
dataset.py Add files via upload Dec 30, 2018
depth-prediction.ipynb
model_utils.py Add files via upload Dec 30, 2018
nn_model.py Add files via upload Dec 30, 2018
plot_utils.py Add files via upload Dec 30, 2018

README.md

Depth-Map-Prediction-from-a-Single-Image-using-a-Multi-Scale-Deep-Network

PyTorch implementation from the papers:
https://cs.nyu.edu/~deigen/depth/depth_nips14.pdf
https://arxiv.org/pdf/1411.4734v4.pdf

Extended model architecture and loss fn to the newer paper.

Model Results:
(image, ground truth depth, model prediction) alt text

Model Arch:

Loss Fn:

Predictions on test image:

Pros:

  • Can detect object boundaries well, due to added image gradient component in the newer loss fn.
  • Prediction quality is decent considering from single image

Cons:

  • Model produces depthmaps at lower resolution (320x240)
  • Depthmaps lack clarity
  • Model is really large, ~900MB, inference time is ~2s for a mini-batch of 8 (640x480) images

Model weights: https://oregonstate.box.com/s/p3lbkgiwufg9rxfgx53c4svnzz2lz9av
NYU Depth Datasets: https://cs.nyu.edu/~silberman/datasets/

You can’t perform that action at this time.